C3SR ushers in new age of computing

The next few years of AI research will shape the future of human life—and some of the leading research in the area is being done right here at Illinois. The Center for Cognitive Computing Systems Research (C³SR) is a joint venture between IBM and the University of Illinois at Urbana-Champaign’s Coordinated Science Laboratory, and its research has been attracting international attention.

AI applications can help humans learn better, work better, and live better, but only if better methods are developed for turning massive unstructured and structured data sets into AI solutions, and for deploying those solutions efficiently and economically. C³SR was created in 2016 with the mission of innovating future heterogenous computing systems, more productive software programs, and methodologies to support the large-scale deployment of a variety of game-changing AI applications.

“In the future, computing is going to be more heterogenous (having more than one processor), and the AI solution deployment will be hybrid with both cloud and edge c

omputing being equally important,” said Wen-mei Hwu, CSL Professor, center co-director, and AMD Jerry Sanders Chair of Electrical and Computer Engineering. “We are researching and building the right software and hardware stacks to support that kind of computing and to make developing AI solutions for that kind of computing much more productive.”

C³SR has achieved success on multiple fronts, ranging from state-of-the-art research to enrichment of Illinois students’ university experiences, while its technologies are increasingly impacting the broader research community. Some of the center’s efforts promise to change the landscape of AI systems research and product development over the coming years.

Research
The C³SR team conducts research on a continuum of hardware to software solutions, with the idea that future AI systems can best be optimized only through full-stack co-innovation. Software research helps bridge the gap between AI solution developers and AI hardware designers, as the two communities speak different languages and optimize for different metrics. The software developed by the group helps solution developers fully utilize the underlying computing infrastructure, while AI hardware research can create the right kind of computing infrastructure to support large-scale deployment of various AI solutions.

“The research in challenging AI solutions helps us better understand the design patterns of complex AI solutions and the algorithm, computation, storage capacity, and bandwidth requirements for the underlying infrastructure,” said Jinjun Xiong, center co-director and program director for IBM Cognitive Computing Systems Research at the Thomas J. Watson Research Center. “The C³SR team is currently exploring a number of different AI solutions.”

One of the most challenging and futuristic AI solutions produced through the partnership, the Creative Experiential Learning Advisor (CELA), complements traditional lecture/examination-based learning by actively engaging learners in efforts to understand learners’ personalized needs. Ultimately, it helps instill and reinforce understanding. The CELA work has encompassed some of the most fundamental AI science research topics, ranging from natural language understanding and video and image understanding to computational creativity.

Another AI solution from the C³SR team is an automated paper reviewer matching system. It helps conference program chairs and journal editors automatically identify appropriate reviewers for paper submissions while accounting for various constraints, such as conflicts of interest, workloads, and the ever-evolving research interests of reviewers.

Because the C³SR team has gained deep understanding of the computing infrastructure needed for AI solutions, it can now conduct more focused AI hardware systems research. One project is on Near Memory Acceleration (NMA). With NMA, data are not moved to computing units; rather, computation is brought close to where data are located. Benefits include improved data processing speed, reduced data movement bandwidth, and high energy efficiency, which are particularly important for AI solutions that require processing of large amounts of data.

The C³SR team’s work has been recognized internationally, with researchers taking home three first-place awards in the Human Parsing Challenge at the 2018 Conference on Computer Vision and Pattern Recognition; two first-place awards in the System Design Contest at the 2019 Design Automation Conference; and first place in the NVIDIA AI City Challenge, among other top-five finishes in the last three years. Overall, the group has published 126 papers, most at top-tier AI and systems conferences, and has won three Best Paper Awards, three Best Poster Awards, and two nominations for Best Paper Awards.

“As far as the center is concerned, the biggest benefit of these awards is the elevation of our researchers’ future career opportunities,” said Hwu. “We love to showcase all these awards as part of the achievements of the center, but the biggest value is the career development of students, faculty and IBM collaborators.”

Education
Training of Illinois students in the latest AI science and technology is a big part of the C³SR collaboration. Students can get involved not only by conducting advanced AI research that addresses practical industrial needs, but also by receiving valuable mentoring from experienced IBM researchers on a daily basis.

C³SR’s newest education effort is its partnership with the Illinois Scholars Undergraduate Research Program (ISUR). In Fall 2019, the two entities introduced the Undergraduate Research in Artificial Intelligence (URAI) program on campus. URAI allows undergraduates to work with C³SR faculty and graduate students on AI research, enabling the undergraduates to contribute to new technology, participate in AI competitions, co-author publications, and receive scholarships.

The C³SR faculty are seeking new ways to enrich students’ learning experiences. One recent endeavor, Alchemy, aims to provide a new learning experience for undergraduate and Master’s students who want intensive research experiences. Led by CSL/C³SR faculty member Sanjay Patel, Alchemy allows students to work together to develop research solutions to real-world problems. The goal is for students to develop new technologies with which they can start their own companies, or that an existing company might acquire. The first cohort of Alchemy students are nearing the end of their commitment, and two projects have already attracted interest from potential buyers.

Internships at IBM offer C³SR students another way to gain experience. Since the center’s inception, IBM has hired more than 20 of its students as interns. While each internship officially lasts for one summer, students’ projects often continue long after their internships are over, sometimes at IBM, if a graduate gets hired. Students thereby gain research experience and improve their skills, while also giving IBM access to a pipeline of qualified potential employees.

Broader Tech Impact
The C³SR team wants to produce results that truly matter to the world, so has made the center’s technologies accessible to everyone in the form of open-source software and systems. Thus, others can build on the center’s work. It may seem surprising that IBM is funding open-source research that could help its competitors, but the company has its eye on the long-term benefits.

“The best way to help IBM in the future is when technology they developed becomes widely adopted by the industry in general,” said Hwu. “Some of the developments are released as open-source to be a resource that can lift the entire industry IBM belongs in.”

Of course, IBM also benefits directly from the partnership. Graduates who have done internships or other center research are familiar with IBM’s research and operations. If they join IBM, that experience reduces the amount of onboarding and training needed, and gives the graduates an advantage. The collaboration also raises student awareness of IBM technologies.

Coursework is one way IBM technologies get put in front of students. Hwu has integrated IBM’s Power server into his ECE 408 (Applied Parallel Programming) and ECE 508 (Manycore Parallel Algorithms) courses. Working from their own laptops, students can seamlessly utilize the latest IBM server for their homework and projects. In another course, ECE 498 (Internet of Things and Cognitive Computing), CSL/C³SR professor Deming Chen has introduced Node-RED, an open-source IBM project that allows students to easily develop enterprise-scale IoT solutions.

“The deep collaboration between IBM and Illinois, enabled by C³SR, is unique and has far-reaching impact beyond both partners,” said Xiong. “AI’s future will largely depend on open collaboration and innovation.”